A unified Python framework for machine learning with time series, offering scikit-learn compatible tools for forecasting, classification, clustering, and more.
sktime is a Python library that provides a unified framework for machine learning with time series data. It solves the problem of fragmented tooling by offering a consistent interface for tasks like forecasting, classification, clustering, and anomaly detection, all while maintaining compatibility with scikit-learn.
Data scientists, machine learning engineers, and researchers working with time series data who need a comprehensive, interoperable toolkit for analysis and modeling.
Developers choose sktime for its scikit-learn compatibility, unified API across multiple time series tasks, and rich model composition features, which streamline workflows and reduce the need to switch between disparate libraries.
A unified framework for machine learning with time series
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
Provides a consistent interface for forecasting, classification, clustering, and more, reducing the need to learn multiple libraries as emphasized in the key features.
Seamlessly works with scikit-learn for pipelining, tuning, and validation, leveraging a familiar workflow for model building and composition.
Supports advanced operations like pipelining, ensembling, and reduction, enabling complex composite models as shown in the quickstart examples.
Offers extension templates for custom estimators and has active community support through Discord and mentoring programs, fostering collaboration and customization.
Key areas like time series alignment and detection are labeled as experimental or maturing in the README, which may limit reliability for production-critical tasks.
Installation requires specifying dependency sets with pip, and conda lacks this flexibility, leading to potential bloat or missing dependencies for specific use cases.
The unified API and integration layers might introduce performance trade-offs compared to using specialized, optimized libraries for individual time series algorithms.